h a l f b a k e r yWhere life imitates science.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
So all decent art drawing packages have a very useful feature called layers, where you
can draw stuff on top of other stuff, and not destroy that underlay. Conceptually, you
are drawing on a series of parallel planes.
Typically you can add and rearrange layers - they're not assigned a specific
depth.
Which is fine, except that sometimes what you're drawing isn't perpendicular to the
viewing angle, so stuff which is in front in one place is behind in another.
So - we let the user define the distance of each layer, and also optionally to slope them.
This feature would probably be quite useful to the power-user very occasionally, but
very awkward, confusing and hard to learn.
Test case - three mutually overlapping rectangles
https://lc1995.gith.../images/Overlap.png A overlaps B overlaps C overlaps A [Loris, Feb 19 2022]
Local Layering [2009]
http://graphics.cs....cts/local-layering/ Research implementation of a similar idea. [markluffel, Mar 06 2022]
[link]
|
|
//Not sure, but it sounds like youre after something like
Autocad or another 3D modeling package.// |
|
|
No, I'm not after that. 3D modelling is quite a different way
of working.
The idea is that this is a feature which would[1] suit digital
artists. |
|
|
[1] well, theoretically could...[2]
[2]Obviously it almost certainly wouldn't. It would be far to
awkward for anyone to ever learn to use. |
|
|
A bit of CAD functionality built into a drawing
program. Why approximate an oval when you can go
off-plane and draw a circle? |
|
|
What would you want for the render quality of the
original skewed image once your viewpoint were out
of plane? Keeping that civil could minimize system
demand. |
|
|
No, you misunderstand. This is for /drawing/ programs.
Layer operations don't distort the image - you change a layer's
position within the stack without skewing or rescaling it. This
merely affects what's 'on top'. |
|
|
Its more like making a mask on a layer and defining the masked and unmasked areas as having different rankings in the heirarchy? |
|
|
I fully understand the idea and think the concept is
of interest. Being able to tilt a layer when viewed
from a cross section viewpoint would acheive this
but varying the transparency range from left to
right or top to bottom would acheive the same
result. [+] |
|
|
////varying the transparency range////
//Explains it well to this non-artist, thanks!//
I'm sure it explains /something/, but it isn't this, because
for any particular
pixel you want either one layer or the other to be in front;
there's no gradual sweep across
the image. |
|
|
//Isnt there a contradiction there? You want to assign
specific depths to layers with some
not parallel to the
others (and not perpendicular to the viewer) but not a 3D
drawing program?// |
|
|
I can see how you might think that, but no, not really. |
|
|
The current state of affairs in graphic drawing programs
like Inkscape, Aseprite, photoshop
and many others is
that you can draw images on one or more layers.
You might think of these as perfectly transparent, infinitely
thin sheets of paper.
Assuming it isn't obscured by anything else (and excluding
layer-specific effects), it doesn't
matter which layer
an image is on, it'll look just the same. |
|
|
One older way of animating things was to draw a
background, then layer on characters and
other foreground on
sheets of cellophane. This has the advantage that you don't
need to redraw the entire scene
for every frame -
just those elements which have changed. |
|
|
So it is with a drawing program. If you decide that you've
made a mistake, you haven't
destroyed whatever it
covers up - because that's still there, just hidden. So you
can move or redraw that element,
and everything is
fine. |
|
|
Which is all well and good. But when things cross-over in
the viewing field, you can't cleanly
separate them to
different layers; one layer is completely in front of the
other. So you end up having to erase part of the top-
most layer image so that the one behind can show through.
Obviously, this means that
you can't adjust either
of the objects without having your carefully aligned 'holes'
misaligned. |
|
|
So what you want is for one side of a layer to be at a higher
position within the stack than
the other. |
|
|
You lost me at //when things cross-over in the viewing field//. The whole point of layers is to allow one to obscure the other. |
|
|
//You lost me at //when things cross-over in the viewing
field//. The whole point of layers is to allow one to obscure
the other.// |
|
|
Sure. But not everything is a convex hull. If everything is
well-separated, there's no problem. But sometimes
different parts of a thing can be in front of and behind
another. |
|
|
Example - imagine you have the front of a house, with an
open doorway in the middle. Imagine viewing it obliquely,
with a character standing just inside the house. You can
only see part of the character, because the near side of the
doorway occludes them. However, they cover the far side
of the doorframe.
This is a deliberately simple example, and the
straightforward solution for this sort of thing is be to split
the house to two layers, bracketing the character's layer.
However, there are more complicated (and reasonably
common) cases where you don't really want to be splitting
off parts of e.g. interacting characters. |
|
|
//[xenzag] provided an explanation fitting your idea into
the drawing layer paradigm.// |
|
|
xenzag mentioned varying the transparency, I think
wouldn't work, but it seemed to make sense to you - so why
don't you explain how that works to me? |
|
|
//Your explanation of a person in a house, viewed obliquely
through a doorway, is a reasonable example of what
someone might draw in a 3D modeling program. But thats
not what youre after either.// |
|
|
I'm going for providing some of the nuanced ability of
managing objects in 3d space, which 3D modelling provides
- to artists who want to draw highly complex and
interesting shapes, like people, creatures, plants and
artistic geometry - without the pain of 3D modelling (and
texturing) complex shapes and geometry. |
|
|
I'll try and dig up a simple example later. |
|
|
//A layer would appear on top where it was opaque to
what was behind it, but hidden where it was transparent
the underlayer would be showing.// |
|
|
I think that is an alternative way of doing it in some
programs - you generate a mask layer by copying one of your
layers, process the colours and reconfigure the layer, then erase areas of that to
show through the original layer to the one behind. It's a bit hacky, and rather
awkward though. |
|
|
I found a demonstration case, linked above. |
|
|
Incidentally, once the concept is clear, I'll move on to
specifying layers with their own z-buffering (basically, a
per-pixel heightmap, promoting the contents above higher
layers), which are more powerful, complex, and supremely
functional. |
|
|
"Sloping" a layer through others is all very well, but that
only gives a linear "cross-over". Based on your "doorway"
example, you are wanting a much more complex boundary,
which would need to be precisely define at each & every
cross-over. So, as others have said, it would be virtually
identical in practice to simply duplicate & erase parts of
the top copy of the layer. (The bottom one can remain
whole as it is obscured by the top.)
Can layers be "linked", so the top one retains all the data
from the "whole" lower version (so any edits to the lower
are reflected in the upper), but only shows part of it,
so future edits can "un-erase" if required (useful for
animation, for example)? |
|
|
//"Sloping" a layer through others is all very well, but that only gives a linear "cross-over". Based on your "doorway" example,
you are wanting a much more complex boundary, which would need to be precisely define at each & every cross-over.// |
|
|
Yes, idea as described in the initial description would be limited to a linear crossover with each other layer. Of course, more
complex geometries are possible... but I figured I'd describe the simplest case first.
However, for most purposes I think this would be sufficient. I disagree that one needs to precisely define each crossover -
they need to be within a tolerance of the respective object edges. It's actually erasing/masking which needs precise work.
The doorway example is a very simple vertical crossing. The objective is to occlude the character by the near edge of the
doorframe. This is a single positioning step, and just has to be within the visible width of the door - easily managed. Splitting
the house between layers requires the same precision, but has the disadvantage that now the house can't be managed as a
single entity. The other alternatives discussed (masking layer on character; erase character) require pixel-level precision. |
|
|
//So, as others have said, it would be virtually identical in practice to simply duplicate & erase parts of the top copy of the
layer. (The bottom one can remain whole as it is obscured by the top.)// |
|
|
Identical results in practice being identical end results? Sure. However in terms of workflow - not at all! Having to erase part
of an object so that another object can show through /does/ have to be very precise - and in general it can't be easily
adjusted. Adjust something a few pixels? Do all your masking again.
It's the incredible awkwardness of this which this proposal aims to address. |
|
|
//Can layers be "linked", so the top one retains all the data from the "whole" lower version (so any edits to the lower are
reflected in the upper), but only shows part of it, so future edits can "un-erase" if required (useful for animation, for
example)?//
|
|
|
I can't speak for everything, but I think in Photoshop this would be the mask layer approach. I'm not an artist and
haven't used it myself - from what I've heard it's expensive and buggy in equal measure. I had a quick look in Krita... and I think the answer might be
yes, but to be honest it looked complicated enough that I couldn't be bothered to figure it out.
But of course the downside is that your edits to that mask layer are now work which isn't itself preservable. Could you could have a mask layer for your
mask layer? Maybe, but your file is ballooning in size.
In practice, for complicated interweaves, I think it's generally going to be easier to avoid masking; clone the appropriate parts of your rear layer and paste
them in front of your front-most layer. This keeps you one good copy of all your character/objects. Disregarding filesize growth (which isn't a big issue for
most artists at the moment), the problem with this is a loss of clarity. It's hard enough to draw stuff on the right layer at the best of times, now if you want to
adjust something you need to remember that you've got stuff in front of your main character layer(s), remove them, work with your rear character -
hiding/revealing/partially hiding the front character as appropriate, then remember to re-apply all those sticky-out bits. Fairly error-prone, unless you're
extremely methodical. |
|
|
[Loris] If the art on the layers is simple enough to do
areas of layer distancing, wouldn't masking work? For
complex art, the lines/areas of pixels still has to have
that precision extra layer distancing drawn for each
pixel , for each time the character/object scene
relationship changes. Seems a small window of help. |
|
|
Sketchup (3D) has planes where pictures can be
placed. These planes can be placed relative to each
other in any orientation wanted in 3d space. A camera
can view from any position. |
|
|
//If the art on the layers is simple enough to do areas of layer distancing,
wouldn't masking work? For complex art, the lines/areas of pixels still has to
have that precision extra layer distancing drawn for each pixel , for each time
the character/object scene relationship changes. Seems a small window of
help.// |
|
|
It's not really a question of whether something /can/ work. Masking can
always work. Just drawing everything on a single layer can always work (it's
just a pain). The question is how easy it is to do. |
|
|
Yes, sloping layers won't be useful for every case. If the problem is of the
"character holding a thing" variety - where there is no visible space between
the two objects - then it won't be better than masking. I'm not suggesting
/replacing/ masking, so you could still do that.
Without checking, I assume complex editors like photoshop let you optionally
move multiple layers (such as an image and masking layer) together. That's
not theoretically hard. It won't help every case, though. Similarly, it's not hard
to imagine being able to toggle movement of an image with its layer-slope
characteristics as well as separately. |
|
|
If you imagine the full-fat "z-buffer" version of this concept - where every
point in an image can be raised to a height above the base, would that be
useful? Well, yes of course... but would it be better than just masking layers? That
is, 'better' in the sense - would it be easier for an artist to do this than using
the current standard of mask layers, assuming all convenience features?
Thinking about it, I think the answer is almost certainly yes. If you had two or
more tightly interwoven images, I think generating a rough heightmap for
one of them would be straightforward, and then you'd be able to pixel-level
adjust relative positioning without having to regenerate anything. That's much less true for masking
or partial-image copying approaches. |
|
|
Thank you markliffel for the link, it's an interesting and I think
rather intuitive way of doing it. Also the description and
examples are clear, so if you still don't know what I'm going on
about you could do worse than have a look at it.
I'd like to welcome you to the halfbakery - but I see you've had
an account for some time, so congratulations on de-lurking. |
|
| |